miRTissue: a web application for the analysis of miRNA-target interactions in human tissues

Bioinformatics

One of the main focuses of translational medicine is the comprehension of molecular mechanisms that characterise the cellular behaviour of complex human diseases [1]. microRNAs (miRNAs) are “master regulators” of gene expression [2], and several pieces of evidence show their involvement in physiological and pathological processes by interaction with target genes [3, 4]. In cancer, one of the main relevant aspects of miRNAs is that they can act as “oncogenes” or “tumor suppressor” genes depending on which target they bind and the cellular environment [58]. Moreover, there are several in vitro functional studies evidencing this dual role of miRNAs in tumorigenesis. As an example, the over-expression of mir-17-92 cluster, considered as an oncogene, is related to lymphoproliferative malignancies. The over-expression of let-7 tumor-suppressor is, in turn, related to reduced tumor burden [9]. Also in breast cancer miRNAs are considered promising molecular “biomarkers” as their profiling can be associated with different breast cancer subtypes, helping to differentiate patients by the different response to therapies, and improving this way the clinical management of patients [10, 11].

miRNAs act on target gene through the interaction with a target sequence within RNA messenger (mRNA). The interaction occurs mainly through the recognition and imperfect binding of 3’ untranslated region (UTR) of mRNA, and also (less frequently reported) of the 5’ UTR and coding sequences (CDS) regions [1214]. The main portion of miRNA sequence interacting with mRNA target is called “seed sequence”, and it is 6-8 nucleotide long [15]. However, the remaining part of the small non-coding RNA contributes to making miRNA-target bond stable. Depending on the bond strength between two interacting molecules and consequently on the degree of homology with the target site, miRNA-target interaction can lead to two different mechanisms [16]: (1) miRNA can lead to mRNA cleavage, and consequently its degradation (Fig. 1a), or (2) miRNA can block protein translation process (Fig. 1b). The second mechanism is detectable by looking at the abundance levels of proteins, and not only of transcripts. In other words, the abundance of proteins in tissue can play a fundamental role in the miRNA-target analysis.

Fig. 1

Mechanism of miRNA-target interaction. a mRNA degradation and b translation inhibition schema. a mRNA degradation: after binding miRISC complex, there is a recruitment of a deadenylase complex (CAF1-CCR4-NOT) acting on 3’UTR region of mRNA. Poli A tail is then removed [47, 48]. After deadenylation, decapping of 5’UTR may occur through a synergistic action of different protein factors (DCP1, DCP2, DDX6, EDC4) [47]. Finally 5’-3’ exonucleases lead mRNA degradation [49]. b Translation inhibition: Translation repression is due to miRNA intervention in different steps of translation. AGO protein has been showed to compete with 5’capping protein factors [50], blocking translation at initiation step. Other mechanisms of miRNA action involve elongation step, causing a premature protein termination [51]

Bioinformatics services and tools, such as databases (db) of validated miRNA-target interactions as well as miRNA-target predictors [17, 18] provided a significant contribution to the investigation of miRNA-target interactions. In this topic, one of the main obstacles is to achieve a high grade of specificity and sensitivity [19]. To overcome this issue, some predictors have introduced experimentally validated interactions. One example is DIANA-MicroT-CDS predictor [20]. It provides miRNA-target interactions both predicted and validated by in vitro experiments, through the support of TarBase db [21]. The latter is a manually curated miRNA-target database, experimentally supported and it includes targets derived from high throughput experiments. Although validations improve the results of interactions, it still lacks information about tissue-specificity because no hint is given about which tissue is more likely to exhibit a predicted and/or validated interaction. MiRWalk system [22] also provides predicted and validated miRNA-target interactions, but it also lacks tissue information. Although the introduction of experimental approaches reduced the number of false positive predictions, that biological information is still quite incomplete, as not all tissues express the same molecules (miRNA and mRNA) at the same time. Moreover, there is the need to have more information on interactions between RNA molecules in a specific case study, and in one particular tissue context.

As an example, we report the case of a miRNA that is over-expressed in a specific tissue type or a certain cellular condition, but its predicted target is not expressed in the same tissue. The produced effect on the cell’s phenotype of such a miRNA can be expected to be rather small. Of course, to overcome this evidence, a validation approach is needed. In particular, the inclusion of expression values in such bioinformatics methods could help to define the molecular interactions between the discussed molecules better. Indeed, recently few miRNA-target prediction methods began to integrate the expression levels of both RNA molecules [23, 24]. For example, the ComiR algorithm [25] uses user-provided miRNA expression levels together with thermodynamic modelling and machine learning techniques to make more accurate predictions, but no straight miRNA-target couples can be evidenced and analysed. miRTarBase [24] is an experimentally validated miRNA-target interaction db, which provides tissue information about miRNA-target interactions, by considering Pearson correlation between miRNA and gene expression profiles. Another exciting predictor is miRGator [26]. It uses both validated interactions and expression values to characterise interactions with regards to specific tumor tissues, by means once again of Pearson correlation. All the above-described bioinformatics tools lack, however, data regarding protein abundance. That kind of data, in fact, is needed to predict interactions causing protein translation inhibition and that, therefore, can not be predicted only considering miRNA and mRNA expression profiles.

In this paper, we present miRTissue, a web application that can provide, for 15 types of human tissues (both tumor and normal), the type of miRNA-target interaction of those validated pairs. To do that, miRTissue exploits the statistical correlation among expression profiles of miRNAs, genes and proteins. In this way, it is possible to have lesser false positive, removing validated interaction not statistically related to a specific tissue, and at the same time to have more sensitive and specific responses. These features can provide a close view of the miRNA status of cells, tissues or organisms. Moreover, miRTissue is the first service that includes protein expression values for the analysis of miRNA-target interactions, in order to provide a novel insight into the type of interactions.

Articles You May Like

Triplet superconductivity demonstrated under high pressure
Why language technology can’t handle Game of Thrones (yet)
Defying the laws of physics? Engineers demonstrate bubbles of sand
BRB-seq: The quick and cheaper future of RNA sequencing
Data mining digs up hidden clues to major California earthquake triggers

Leave a Reply

Your email address will not be published. Required fields are marked *